Abstract: A time series model of wind direction is presented based on stochastic Markov chain model in this paper, in which the 16 wind-direction partitioning method in meteorology is used as the basis for partitioning the state spaces of stochastic Markov chain model. Model verification shows that the generation of time series of wind direction by stochastic Markov chain model can retains such better statistical characteristics as distribution characteristic of original data, autocorrelation coefficient and power spectrum characteristics. It is proved that it is feasible that the time series of wind direction in arbitrary time length is generated based on the stochastic Markov chain model, and the model has better prediction effect in forecasting wind direction by introducing threshold correction.